The ISME Journal (2015) 9, 683695

& 2015 International Society for Microbial Ecology All rights reserved 1751-7362/15
www.nature.com/ismej

ORIGINAL ARTICLE
Bacterial assembly and temporal dynamics in
activated sludge of a full-scale municipal wastewater
treatment plant
Feng Ju and Tong Zhang
Environmental Biotechnology Laboratory, Department of Civil Engineering, The University of Hong Kong,
Hong Kong SAR, China

Understanding environmental and biological influences on the dynamics of microbial communities

has received great attention in microbial ecology. Here, utilizing large time-series 16S rRNA gene
data, we show that in activated sludge of an environmentally important municipal wastewater
treatment plant, 5-year temporal dynamics of bacterial community shows no significant seasonal
succession, but is consistent with deterministic assemblage by taxonomic relatedness. Biological
interactions are dominant drivers in determining the bacterial community assembly, whereas
environmental conditions (mainly sludge retention time and inorganic nitrogen) partially explain
phylogenetic and quantitative variances and indirectly influence bacterial assembly. We demon-
strate a correlation-based statistical method to integrate bacterial association networks with their
taxonomic affiliations to predict community-wide co-occurrence and co-exclusion patterns. The
results show that although taxonomically closely related bacteria tend to positively co-occur (for
example, out of a cooperative relationship), negative co-excluding correlations are deterministically
observed between taxonomically less related species, probably implicating roles of competition in
determining bacterial assembly. Overall, disclosures of the positive and negative speciesspecies
relations will improve our understanding of ecological niches occupied by unknown species and
help to predict their biological functions in ecosystems.
The ISME Journal (2015) 9, 683695; doi:10.1038/ismej.2014.162; published online 2 September 2014

Introduction In the past two decades, great efforts have been

As the most popular biological wastewater treatment
application worldwide, activated sludge has been in
use for exactly a century to treat a large variety of
municipal and industrial wastewaters to protect our
environment and human health (Wagner and Loy,
2002; Seviour and Nielsen, 2010). Activated sludge
is a unique artificial microbial ecosystem with high
diversity (over 700 genera and thousands of opera-
tional taxonomic units (OTUs)) (Zhang et al., 2012)
and with high biomass concentration (generally
210 g l  1) (Grady et al., 2011). The highly diverse
bacterial communities in this engineered ecosystem
efficiently aggregate themselves in the hetero-
geneous structure of activated sludge flocs to
guarantee stable and good performance of biological
wastewater treatment (Daims et al., 2006; Nielsen
et al., 2012; Zhang et al., 2012; Ju et al., 2013a).
Correspondence: T Zhang, Environmental Biotechnology Laboratory,
The University of Hong Kong, Pokfulam Road, Hong Kong,
China.
E-mail: zhangt@hku.hk
Received 12 March 2014; revised 11 May 2014; accepted 17 July
2014; published online 2 September 2014
made to isolate, characterize or quantify functional
microorganisms directly involved in removing
nutrients (nitrogen and phosphorus) (Bond et al.,
1995; Juretschko et al., 1998; Daims et al., 2006),
hydrolyzing and fermenting bacteria (Juretschko
et al., 1998; Xia et al., 2008), floc-forming bacteria
(Shin et al., 1993; Schmid et al., 2003) and
detrimental microorganisms that raise bulking and
foaming problems (Wanner, 1994; Guo and Zhang,
2012) in activated sludge. Despite a rapidly increas-
ing knowledge concerning the biochemical and
ecological characteristics of these key microbes in
wastewater treatment, full-scale activated sludge-
based wastewater treatment plants (WWTPs) with
nutrient removal still suffer from a series of
operational problems, such as process instability
(Eikelboom, 2000), sludge settling problems (Jenkins
et al., 2004) and poor performance in nutrient
removal (Seviour and Nielsen, 2010). Therefore,
more fundamental knowledge regarding the micro-
bial structure is essential to elucidate the biological
mechanisms behind the problems.
Recently, high-throughput culture-independent
sequencing tools, such as 16S rRNA-based pyrose-
quencing, have been widely used to survey and
 Bacterial assembly and temporal dynamics in AS
F Ju and T Zhang
684
improve our understanding of biodiversity in var-
ious municipal or industrial WWTPs (Kwon et al.,
2010; McLellan et al., 2010; Ibarbalz et al., 2013).
However, these descriptive studies have left many
unanswered questions regarding the underlying
speciesspecies interactions and environment
species relations driving bacterial community
assembly and dynamics (Ju et al., 2013b). Further,
the long-term temporal variation in microbial
constituents and interactions in a WWTP over
environmental gradients remains largely unknown
due to either the unavailability of sufficient time-
series samples and long-term physicochemical and
biological monitoring data, or the lack of a powerful
analytical method for mining the huge sequence
data derived from high-throughput sequencing.
In this study, we applied a correlation-based
network analysis, which was based on over
570 000 bacterial 16S rRNA gene sequences
from 58 activated sludge samples collected monthly
from a typical municipal WWTP over 5 years (2007
2012), to explore the long-term bacterial assembly
and temporal speciesspecies associations (SSAs) in
activated sludge. We described a correlation-based
statistical method to integrate bacterial SSA net-
works with their taxonomic affiliations to reveal the
non-random assembly patterns among species. Crea-
tively interpreting OTUs in an SSA network may
help to transit the microbial ecology of activated
sludge from a plain description of microbial
components to a framework of potential microbial
interactions, in which ecological rules guiding the
microbial assembly and functions could be specu-
lated for the encouragement of further validation via
specific experimental designs or directly applied to
guide the system toward an optimized performance.
Moreover, the synchronous and time-lagged correla-
tions between potential influential factors (for
example, plant operational parameters and waste-
water quality) and bacterial species, as well as the
contributions of influential factors on the temporal
variability of biodiversity, were also explored in the
network interface and correlated with the functional
stability of activated sludge, which is a good model
of the artificial microbial ecosystems.
Materials and Methods
Sample collection
The sampling site is a full-scale municipal WWTP
(216 000 m3 day  1) in Shatin, Hong Kong (221230 N
1141110 E), which treats saline domestic sewage
containing B30% seawater. The plant is designed
as an anoxic/oxic (A/O) process for carbon and
nitrogen removal. Activated sludge samples were
collected monthly from the middle of aerobic (oxic)
tank from July 2007 to July 2012. The samples were
fixed on site using an equal volume of 100% (v/v)
ethanol. Then, the fixed samples in 50% ethanol
were immediately delivered to the laboratory and
The ISME Journal
stored in a  20 1C refrigerator. Throughout the
entire sampling period, the plant could effectively
remove 9598% of CBOD (carbonaceous biochem-
ical oxygen demand) in the sewage; however, the
plant usually encountered unstable ammonium
removal (4798%) from every December to the next
March, during which period, severe foaming (level
45, 510 cm, highly stable foams; Seviour and
Nielsen, 2010) of activated sludge was usually
observed (Supplementary Figure S1). Other detailed
information concerning variation of plant opera-
tional parameters and physicochemical conditions,
the sampling dates and treatment performance were
summarized in Supplementary Table S1 and
Supplementary Information S1.
DNA extraction and 454 pyrosequencing
For each activated sludge sample, DNA was first
extracted from 2.0 ml sludge using a FastDNA @
SPIN Kit for Soil (MP Biomedicals, LLC, Illkirch,
France), then the V3-V4 regions (B465 nucleotides)
of the 16S rRNA genes were amplified with 338F
and 802R, and purified PCR amplicons were finally
send out for pyrosequencing (see Supplementary
Information S2 for detailed information). All 16S
rRNA sequences from pyrosequencing have been
deposited into the NCBI short-reads archive data-
base with accession number SRR1154613.
Sequence processing
The raw sequencing data from 454 pyrosequencing
were processed using the QIIME pipeline v 1.7.0
(Caporaso et al., 2010). In brief, the raw sequences
were first quality trimmed into different samples,
denoised by Denoiser (Reeder and Knight, 2010) and
chimera checked using ChimeraSlayer (Quince
et al., 2011) to yield clear reads. Then, the normal-
ization of the clear sequences was conducted by
randomly extracting 10 000 clean sequences from
each sample data set (except for two samples from
March and April, 2011, which have clean sequences
of 9524 and 6257, respectively) to fairly compare all
samples at the same sequencing depth. Next, the
normalized sequences from all samples were clus-
tered into OTUs using the Uclust algorithm at
identity thresholds of 0.90 and 0.97 (Edgar, 2010),
which approximately corresponding to the taxo-
nomic levels of family and species for bacteria,
respectively. Both the final 0.90 and 0.97-OTU tables
consisted of 575 781 clear sequences, which were
distributed into 2192 family-level and 5136 species-
level bacterial OTUs, respectively. Of those OTUs, 861
and 2075, respectively, were represented by at least
five OTUs. Finally, the taxonomic assignment of the
representative sequences was conducted using the
RDP Classifier program (80% confidence level) and
GreenGenes database newly released in May 2013
(McDonald et al., 2012).

Statistical and network analysis
The core diversity analyses (using QIIME v 1.7.0),
BIO-ENV analysis (using PRIMER-E v6 software
(PRIMER-E Ltd, Ivybridge, UK), Spearmans rank
correlation method and a significance test of 99
permutations to determine the combination of
environmental variables that best explain community
patterns) and correlation analysis (using R; Ihaka
and Gentleman, 1996) between a-diversity indices
and 15 environmental variables (Supplementary
Table S1) were described in Supplementary
Information S3.
For the network analysis, we first used extended
Local Similarity Analysis to find the time-
dependent correlations between species-level
OTUs and environmental variables (Ruan et al.,
2006; Xia et al., 2011). The Local Similarity
Analysis calculates synchronous and time-delayed
correlations based on the normalized ranked
data and produces correlation coefficients
that are analogous to a Spearmans ranked
correlation (Ruan et al., 2006). Then, we used
Cytopscape v2.8.3 (Shannon et al., 2003) for
network visualization and topological analysis, as
described in Supplementary Information S3.
Here, we developed a python script to check
statistically the observed (O) and random
incidences (R) of bacterial co-occurrence and
co-exclusion (Supplementary Information S5).
The degree of the lack of agreement between
O and R (O/R ratio; Supplementary Table S7;
Supplementary information S7) is used as a bench-
mark for checking non-random assembly patterns in
complex bacterial communities.
Figure 1 Three-dimensional principal coordinate analysis (PCoA) plot showing the bacterial community difference of the 5-year
activated sludge samples. The analysis was performed using the abundance matrix of family-level (0.90 similarity) OTUs in different
samples, and pairwise community distances were determined using the weighted UniFrac algorithm.
Bacterial assembly and temporal dynamics in AS
F Ju and T Zhang
685
Results
Monthly, seasonal and inter-annual variability of
bacterial structure
The seasonal succession of bacterial communities
over various environmental gradients has been
widely observed in many natural ecosystems,
including soil (Lipson and Schmidt, 2004), oceans
(Gilbert et al., 2011), lakes (Eiler et al., 2011;
Paver et al., 2013). However, in an artificially
controlled, half-close engineered biological waste-
water treatment system, such as activated sludge,
whether the dynamics of the bacterial community
structure still follow a seasonal succession remains
to be explored. Here, we compared the monthly,
seasonal and inter-annual variations in bacterial
diversity and abundance between time-series acti-
vated sludge samples using weighted UniFrac
distances (considering both species phylogeny and
abundance). Overall, temporal changes in the
phylogenetic composition and abundance of
family-level OTUs were quite high across the 5-year
sampling period (Figure 1; see Supplementary
Figure S3 for similar trends at the species level).
Unlike the aforementioned natural ecosystems,
there was no obvious seasonal succession of bacter-
ial communities in the artificially controlled bio-
technical ecosystem (for example, activated sludge)
because samples collected from the same or adjacent
months in different years were hardly clustered
together. Instead, an annual shift in the communities
proceeded in large leaps, that is, from ellipse I
(20072008), via ellipse II (20092010) and finally
into ellipse III (20112012), at particular time slots,
The ISME Journal
 Bacterial assembly and temporal dynamics in AS
F Ju and T Zhang
686
which were usually found in winter months from
around December to the next February (as illustrated
by the white arrows within one or across different
ellipses). This period was exactly the time when
severe filamentous foaming was usually observed at
the surface of the aeration tank (AT) of Shatin
WWTP (also as indicated by the prevalence of
bacterial OTUs potentially related to bulking and
foaming; Supplementary Figure S6), accompanied
by higher levels of NH3-N (Supplementary
Figure S2a) and relatively lower Z(NH3-N) (that
is, removal efficiency of NH3-N; Supplementary
Figure S2d; Supplementary Table S1) in the AT.
Strikingly, the largest monthly variability was
observed between November 2010 and February
2011, exactly when low F/M (Supplementary
Figure S2b) bulking, together with filamentous
foaming, occurred in the AT, indicating that
these detrimental events greatly affect both the
phylogenic and quantitative profiles of bacterial
communities in activated sludge.
Environmental influences on bacterial diversity and
abundance
Correlation analysis between bacterial a-diversity
matrices and physicochemical and operational vari-
ables showed that on the one hand, bacterial
a-diversity in activated sludge was most closely
positively correlated with sludge retention time
(SRT) (coefficients of 0.600.03) and mixed liquor
suspended solids (0.550.02), followed by NO3-N
(0.550.03) and Z(NH3-N) (0.540.10) (cluster I,
Supplementary Figure S4; Supplementary Table S2);
on the other hand, NO2-N-AT (NO2-N concentration
in the AT,  0.610.05), influent CBOD
(  0.570.03), F/M (  0.510.03) and NH3-N-AT
(  0.510.08) showed significantly negative correla-
tions with the a-diversity matrices (Supplementary
Table S2 and cluster III, Supplementary Figure S4).
Additionally, other physicochemical (for example,
temperature, salinity and pH) and operational
parameters (hydraulic retention time and dissolved
oxygen) had either little or no statistically signifi-
cant (P-value40.05) correlation with the a-diversity
matrices (cluster III, Supplementary Figure S4),
indicating that these parameters may have little
impact on activated sludge bacterial diversity.
Once we demonstrated that bacterial a-diversity
in activated sludge could be influenced by environ-
mental variables, a BIO-ENV trend correlation
analysis was conducted to identify which combina-
tion of variables best explained changes in the
bacterial abundance and in the diversity over time
(that is, b-diversity). Supplementary Table S3 shows
that operational parameters, in general, explained
much better variation in the change in the bacterial
structure than did physiochemical variables in
the influent (for example, NH3-N and CBOD) or in the
AT (for example, temperature, pH and salinity). The
variables that best explained the weighted Unifrac
The ISME Journal
distance between all pairs of 0.97 or 0.90-OTUs were
identical and included SRT and F/M (BEST: 0.97-
OTUs, rho 0.483; P-value 0.01; 0.90-OTUs,
rho 0.471, P-value 0.01, respectively). Both level
OTUs still correlated best with these two opera-
tional parameters when unweighted (without
considering abundance) Unifrac distance matrices
were used instead, except for the incorporation of
some physiochemical parameters, which included
influent NH3-N and CBOD and NO3-N in the AT.
Defining the bacterial community by frequency and
functionality
Partitioning ecological communities by their abun-
dance and by their occurring frequency facilitate the
exploration of the core and satellite species in many
temporal or spatial scale data sets. In general,
satellite species were typically transient and low
in abundance, whereas core species were persistent
in a given habitat and high in abundance (van der
Gast et al., 2010). On the basis of the occurrence
frequency, we divided the bacterial community of
activated sludge into the following three arbitrarily
defined ecological categories: persistent (X80% of
months), intermittent (2080% exclusive) and tran-
sient (p20%) OTUs (Figure 2). Overall, positive
relations between the mean abundance and occur-
rence frequency have been observed, which were
best fitted using the following exponential equation:
Y 0.0013e0.0932X (R2 0.82) (Figure 2a). This
revealed that persistent OTUs were generally more
abundant than intermittent and transient OTUs,
although the former ecological categories included a
much lower number of OTUs (Figure 2b). Specifi-
cally, persistent OTUs merely occupied 9.7% of
2075 bacterial OTUs but accounted for 76.6% of all
16S rRNA gene sequences, implicating the existence
of a high proportion of longstanding core species in
activated sludge to sustain its long-term functional
stability. By contrast, transient OTUs composed over
56.0% of all bacterial OTUs but merely occupied a
minor proportion (3.4%) (Figure 2c), suggesting an
extremely high diversity of minority species in
activated sludge.
The taxonomic distribution of persistent, inter-
mittent and transient OTUs was slightly different
from each other (Supplementary Figure S5). Some
bacterial classes, such as Alphaproteobacteria,
Actinobacteria and Nitrospira, tended to be more
persistent, whereas others (for example, Delta-
proteobacteria, most sulfate-reducing bacteria in
this class, which depend on the aerobic/anaerobic
condition) were more transitory over the 5-year
sampling period. Persistent OTUs were primarily
affiliated with 12 classes, such as Alphaproteobacteria,
Actinobacteria, Gammaproteobacteria, Acidimicrobia,
Sphingobacteria and Anaerolineae, which are also
key bacterial groups commonly found in activated
sludge of different municipal WWTPs (Wagner and
Loy, 2002; Sanapareddy et al., 2009; Xia et al., 2010;
 Bacterial assembly and temporal dynamics in AS
F Ju and T Zhang
687

Figure 2 Defining the 5-year activated sludge microbiome at the Shatin WWTP. The mean abundance of OTUs (a) and the number of
OTUs (b) are shown relative to the 0.97-OTUs percentage occurrence (X axis). The occurrence frequency was calculated by dividing the
number of months in which an OTU was detected by the number of total months. The abundance and the number of OTUs of different
occurrence frequencies (Persistent, Intermittent and Transient) are shown in (c). Diversity (d) and mean abundance (e) of potential
functional bacterial groups (in the monthly samples they were detected) of different occurrence frequencies (see Supplementary Table S4
for the list of the functional bacteria).

Zhang et al., 2012). By contrast, intermittent and
transient OTUs included a larger proportion
(3035%, Supplementary Table S4) of populations
from other bacterial classes, such as TM7-1, TM7-3,
Synergistia, Verrucomicrobiae and Chlamydia
(Supplementary Table S4). Noteworthy, the
predominance of Alphaproteobacteria predominated
over Betaproteobacteria is mainly attributed by the
salinity (B1%) of wastewater in Shatin WWTPs, as
we discussed previously (Zhang et al., 2012).
Further comparison with previous studies
(Jenkins et al., 2004; McLellan et al., 2010; Seviour
and Nielsen, 2010; Guo et al., 2013) implicated high
diversity (209 OTUs, Figure 2d) and a considerable
proportion (averaged 25.7% of bacterial 16S rRNA
sequences, Figure 2e) of potentially functional
bacteria (see Supplementary Table S4 for a full
list) in activated sludge, including nitrifying
bacteria (two OTUs of ammonia-oxidizing bacteria
(AOB); four OTUs of nitrite-oxidizing bacteria
(NOB)), phosphate accumulating organisms (four
OTUs), glycogen accumulating organisms (three
OTUs), hydrolyzers (40 OTUs), bulking and
foaming bacteria (BFB, 76 OTUs), denitrifiers
(24 OTUs) and fermentative human-fecal bacteria
(57 OTUs). Although persistent functionalists
(42 OTUs, 20% of the total number of functional
OTUs) were represented by much fewer OTUs than
intermittent (94) and transient functionalists (74),
they accounted for over 70% of the 16S sequences of

The ISME Journal

 Bacterial assembly and temporal dynamics in AS
F Ju and T Zhang
688

Figure 3 Environmentspecies network uncovered synchronous and delayed relations between bacteria and environmental variables in
activated sludge. Only local similarity that were statistically significant (P-value p0.05, Q-value p0.01) and strong (local similarity X0.6
or p  0.6) are shown, resulting in networks composed of 67 nodes and 115 edges. Node label stands for an environmental variable or the
lowest classifiable taxonomic rank (p_, c_, o_, f_ and g_ representing phylum, class, order, family and genus, respectively) of 0.97-OTU,
and node size of each OTU is proportional to its average abundance in the samples it was detected. The line thickness is proportional to
the absolute value of local similarity, and line arrows indicate a 1-month shift/delay in the correlation.

all functional bacteria, implicating the longstanding
co-existence of a core set of functional bacteria in
activated sludge.
Figures 3e and 4f show OTUs related to bulking
and foaming (averaged 12.5% of bacterial 16S rRNA
sequences), which primarily consisted of filamen-
tous Microthrixaceae (averaged 2.5%, 7 OTUs),
Caldilinea in the phylum Chloroflexi (1.5%, 23 OTUs),
hydrophobic Mycobacterium (3.9%, 18 OTUs) and
filamentous, hydrophobic Gordonia (3.8%, 7 OTUs).
Although these notorious and always filamentous
BFB, if present outside the bioflocs, can cause
settling (bulking) and foaming problems and
deteriorate effluent quality, it is believed that BFB-
related filaments are usually presented in
well-behaved activated sludge and have versatile
roles (for example, bioflocs formation; Kragelund
et al., 2007; lipids or oleic acid degradation, Nielsen
et al., 2010) other than being detrimental. Moreover,
the protein hydrolyzers Saprospiraceae (phylum
Bacteroidetes) were highly diverse (40 OTUs,
Figure 2d) and abundant (5.0%, Figure 2e) in
activated sludge. Intriguingly, the poor representa-
tiveness of AOB in activated sludge hardly hindered
NH3-N oxidization (as indicated by the continuously
detected nitrite in the AT; Supplementary Table S1),
most likely justifying previous findings that
Nitrosomonas has high transcription activity in
spite of its low abundance in activated sludge
(Yu and Zhang, 2012). It is also possible that there
are unassigned or unidentified AOB.
Environmentspecies association and SSA
Tracking correlations between microorganisms and
between microorganisms and their surrounding
environments in a network interface provide
insights into microbial interactions, as well as an
awareness of the conditions that favor or disfavor
particular microbes. Restricting the analysis to the
environmentspecies association networks (Figure 3;
Supplementary Figure S7), strong correlations
between variables including NH3-N-AT, NO2-N-AT,
NO3-N-AT and SRT and bacterial taxa were the most
frequent, followed by those correlations between
taxa and other variables, such as mixed liquor
suspended solids, temperature and F/M. Few
significant and no significant, strong correlations
were observed between bacterial taxa and other
variables, including hydraulic retention time,
dissolved oxygen, pH and SCOD influent
(Supplementary Figure S9). Strikingly, variables
with more edges connected to bacterial taxa, that is,
NH3-N-AT, NO2-N-AT, NO3-N-AT and SRT, tend
to have much better correlation with bacterial
a-diversity than those variables with fewer edges
(for example, hydraulic retention time, dissolved
oxygen and pH) (Supplementary Figure S4),
confirming that SRT and inorganic nitrogen in the
AT, compared with other environmental variables,
may much more significantly affect the community
structure.
The mathematical statistics of the environment
species association network (Figure 3) indicate that

The ISME Journal

 Bacterial assembly and temporal dynamics in AS
F Ju and T Zhang
689

Figure 4 Examples of strong and significant correlations between species and environment (a, b) and between species and species (c, d,
e and f) in 58 activated sludge during 5-year sampling period. An local similarity is considered as strong and significant when local
similarity X0.6 or p  0.6, and P-valuep0.05 and Q-valuep0.01. The OTU abundance is calculated as the number of sequences assigned
to each OTU divided by the total number of 16S rRNA gene sequences in that sample. The missing points in the (d) represent OTUs
abundance of 0.

bacterial taxa were primarily connected to SRT and
NO3-N-AT via positive correlations (both synchro-
nous and delayed). Thus, the increase in SRT and
NO3-N concentrations may promote the accumula-
tion of many bacterial OTUs, such as seven
Rhizobiales-affiliated OTUs (f_Hyphomicrobiaceae,
g_Bauldia, g_Bradyrhizobium and f_Rhodobiaceae),
two hydrolyzer-affiliated OTUs (f_Saprospiraceae),
two Chloroflexi-affiliated OTUs (o_mle1-48) and one
NOB-affiliated OTU (g_Nitrospira). In contrast,
negative correlations dominated the correlations
between NO2-N, NH3-N and bacterial OTUs, reveal-
ing that the buildup of NO2-N and NH3-N in the AT
tends to reduce the abundances of certain bacterial
groups. In addition, some bacterial OTUs, such as
those bacterial OTUs affiliated with g_Conexibacter,
o_Solirubrobacterales, f_Saprospiraceae (Figure 4a)
and f_Moraxellaceae, were positively correlated
with NO3-N-AT with delay, on the one hand, and
negatively correlated with NO2-N-AT and/or NH3-N-AT
on the other hand, indicating that these bacteria
thrive when AT is relatively high in nitrate but low
in nitrite and ammonium.
The analysis of the integrated network
(Supplementary Figure S7) composed of positive-
correlated nodes extracted from the environment
species association network and the SSA network
shows that two groups of environmental variables,
that is, (I) NH3-N-AT & NO2-N-AT and (II) SRT and
NO3-N, exerted considerable impacts on the overall
co-occurrence patterns of the bacterial community,
but to different degrees. Topological partitioning
shows that the network could be divided into two
large clusters (or modules). In the upper cluster,
NH3-N-AT and NO2-N-AT were connected to only a
small proportion of OTUs (12 nodes) on the right

The ISME Journal

 Bacterial assembly and temporal dynamics in AS
690
corner of the cluster, reflecting their limited influ-
ence on the cluster structure. By contrast, in the
lower cluster, SRT and NO3-N were correlated with
more OTUs (28 nodes), many of which are hub
OTUs (nodes with a high number of connections,
also known as high degree nodes) in the center of
the cluster (comparing Supplementary Table S6
against Supplementary Figure S7). Via hubs, the
impacts of these parameters could spread rapidly to
reach neighboring OTUs in other parts of the cluster.
Further topological and taxonomic comparison
lends interesting and novel insights into the com-
munity structure. For instance, almost all OTUs of
three bacterial phyla, including TM7, Chloroflexi
(classes Anaerolineae and Thermomicrobia)
and Actinobacteria (classes Actinobacteria and
Acidimicrobiia), only occurred in the relatively
loosely (70 OTU nodes, 441 edges) connected upper
cluster, whereas almost all OTUs of two classes,
Sphingobacteria and Gammaproteobacteria, were
only found in the densely connected (70 OTU
nodes, 441 edges) lower cluster.
Preferential attachment of bacterial nodes revealed
deterministic co-occurrence and co-exclusion patterns
We constructed correlation-based SSA networks
and used a statistical method to test both the
co-occurrence and co-exclusion patterns between
bacterial communities. The resulting entire SSA
network (local similarity X0.6 or p  0.6) consisted
of 150 nodes and 913 edges (average degree of 12.17
and average shortest path length of 2.866, Table 1;
see Supplementary Figure S10 for the cumulative
degree distribution and exponentially decreased
average shortest path length with increasing node
degree), with 557 positive interactions between 145
OTUs compared with 356 negative interactions
between 117 OTUs (Table 1). The higher clustering
coefficients of the entire and positive SSA networks,
compared with Erdos-Reyni random networks, with
small characteristic shortest path lengths (Table 1),
similar to random graphs, suggest that the network
has small-world properties, that is, nodes that are
highly interconnected (clustered) more than would
be expected by chance alone. By contrast, the
Table 1 Comparison of topological properties of speciesspecies association (SSA) networks of activated sludge with their
corresponding Erdos-Reyni random networks of identical size
SSAa Avg. clustering Avg. shortest path
coefficient length
Whole 0.466 2.866
Random 0.0830.00 2.2620.01
Positive 0.477 4.615
Random 0.0530.00 2.6670.00
Negative 0 3.491
Random 0.0530.00 2.8300.00
a
Whole SSA: 150 nodes, 913 edges; Positive SSA: 145 nodes, 557 edges; Negative SSA: 117 nodes, 356 edge.
The ISME Journal
F Ju and T Zhang
negative SSA, which reflects speciesspecies exclu-
sion patterns, tends to be unclustered (an average
clustering coefficient of 0) and less modularized
(modularity: 0.457, Table 1; values 40.4 suggest that
the network has a modular structure; Newman
2006), compared with the highly clustered, more
modularized (modularity: 0.586) positive SSA,
revealing distinct characteristics of positive and
negative interactions between species.
The further structural and statistical analysis
showed that OTUs from the same taxa (from phylum
down to the order level) tended to co-occur (positive
correlations, Supplementary Figure S7) and that
OTUs from different taxa tended to co-exclude
(negative correlation, Supplementary Figure S8)
more than would be expected by chance when
considering taxa frequency and random associa-
tions, although their degrees of co-occurrence or
co-exclusion (as measured by O/R ratio) differed
(Supplementary Tables S7 and S8). On the one
hand, statistical and structural analysis of the
positive SSA network showed that OTUs within
two orders, that is, Rhizobiales (O/R 4.3) and
Rhodobacterales (O/R 2.0, family Rhodobacteraceae),
and four classes, that is, Sphingobacteria (O/R 5.1,
primarily family Saprospiraceae, Supplementary
Figure S11b), Anaerolineae (O/R 3.6; Figure 4c;
Supplementary Figure S11a), Gammaproteobacteria
(O/R 4.3) and Betaproteobacteria (O/R 2.5),
tended to co-occur more than would be expected
by chance (Supplementary Table S6). On the other
hand, the statistical analysis of the negative SSA
network demonstrated that OTUs from different
taxa, (I) Anaerolineae and Rhodobacterales
(O/R 2.5, Supplementary Figure S11a), (II)
Anaerolineae and Betaproteobacteria (particularly
family Xanthomonadaceae) (O/R 2.7, Supplementary
Figure S11a), (III) Flavobacteria and Thermomicrobia
(O/R 7.1) and (IV) Rhizobiales (particularly family
Hyphomicrobiaceae, for example, Figure 4f) and
TM7 (O/R 1.9), tend to co-exclude more than
would be expected by chance (Supplementary
Table S6). Strikingly, apart from the deterministic
patterns of intra-taxon co-occurrence and inter-taxa
co-exclusion, higher incidences of inter-taxa
co-occurrence, more than would be expected by
Network Network Network Modularity Graph Avg.
heterogeneity centralization diameter density degree
0.785 0.176 8 0.4240.002 0.082 12.170
0.2710.022 0.060 4 0.2390.006 0.082 12.170
0.818 0.135 14 0.5860.001 0.053 7.630
0.3310.014 0.052 5 0.3160.004 0.053 7.630
0.732 0.140 9 0.4570.001 0.052 6.085
0.3900.013 0.061 6 0.3570.004 0.052 6.085
 Bacterial assembly and temporal dynamics in AS
F Ju and T Zhang
691

Figure 5 Preferential attachment of bacterial nodes in the speciesspecies association network revealed deterministic bacterial co-
occurrence (solid edges) and co-exclusion (dash edges) patterns in activated sludge. Only correlations that were statistically significant
(P-value p0.05, Q-valuep0.01) and strong (local similarity X0.6 or p  0.6) are shown. Node label stands for the lowest classifiable
taxonomic rank (p_, c_, o_, f_ and g_ representing phylum, class, order, family and genus, respectively) of 0.97-OTU. Nitrite-oxidizing
bacteria Nitrospira tends to co-occur with OTUs of Sphingobacteria, Gammaproteobacteria and Betaproterobacteria (for example
Nitrosomonas, Thauera and Azoarcus), but co-exclude with OTUs of Actinobacteria. The line thickness is proportional to the absolute
value of local similarity. The arrow indicates the time-lagged correlations with arrow pointing to the lagged OTU.

chance, were also observed between OTUs
of different taxa, including (I) Sphingobacteria
(primarily Saprospiraceae) and Gammaproteobacteria
(Supplementary Figure S11b), (II) Anaerolineae and
TM7 (Supplementary Figure S11a), (III) Nitrospira and
Sphingobacteria or Gammaproteobacteria (Figure 5)
and (IV) other pairs of taxa.
Discussion
It has long been assumed that differences in species
abundance in microbial communities reflect changes
in environmental conditions. Although this state-
ment emphasizes the significance of environmental
impacts, it ignores the influences of interior
speciesspecies interactions on the community
assembly. Currently, it remains difficult to predict
which bacteria can co-exist or co-exclude steadily
over temporal gradients of environmental variables,
let alone the cooperative or competitive relations
among these bacteria, causing the artificial and
purposeful manipulation of engineered microorgan-
isms (for example, in biological WWTPs) extremely
challenging. In this study, utilizing large time-series
16S rRNA gene sequencing data, we constructed a
bacterial SSA network consisting of 3899 pairwise
significant SSA correlations (among which 913
correlations are strong, with coefficients X0.6)
connecting 170 species-level OTUs. We find that
although taxonomically closely related bacteria tend
to co-occur out of cooperative relations or a similar
niche preference, co-excluding negative correlations
are usually deterministically observed between
taxonomically less related species, most likely
implicating a role of competition in community
assembly. Moreover, the highly clustered and
modularized structure (also characterized by nodes
connected by many closed triangle or polygonal
loops) of the positive SSA network is completely
different from the unclustered and less modularized
structure of the negative SSA network. This
result indicated that positive interactions (primar-
ily cooperative relations) among bacteria are
usually established by a cluster of multiple highly
interacted species with similar ecological niches,
whereas bacteria are likely to form relatively
simple and open one-to-many or one-to-one
negative interactions (most likely competition)
with one another.

The ISME Journal

 Bacterial assembly and temporal dynamics in AS
F Ju and T Zhang
692
The non-random co-occurrence patterns between
taxonomically closely related bacteria can be
derived from taxa sharing similar ecological niches
or cooperative relations, as noted elsewhere
(Barberan et al., 2011; Ju et al., 2013b). One typical
example is the intra-taxon co-occurrence observed
among Betaproteobacteria AOB, Nitrosomonas and
two other Betaproteobacteria-affiliated denitrifying
bacteria, that is, Thauera and Azoarcus (Figures 4d
and 5), which resulted from a syntrophic relation, in
which nitrite released by the former is utilized by
the latter. Other examples are the strong intra-taxon
co-occurrences evident in the class Anaerolineae
or in the order Sphingobacteria, which are likely
derived from the preference for similar niches, as
supported by the assemblage of all OTUs of
Anaerolineae in the upper module and all OTUs of
Sphingobacteria in the lower module (Supplementary
Figure S7). Overall, the deterministic intra-taxon
co-occurrence patterns evident between taxonomi-
cally closely related species were in agreement with
the widespread ecological phenomenon of phylo-
genetic clustering (that is, co-occurring species
being more closely related than would be expected Figure 6 Schematic diagram of potential bacterial interactions
by chance), which most likely implicate the impor- among autotrophic and heterotrophic bacteria in a nitrogen-
tance of environmental filtering and niche differ- removal activated sludge treatment plant. Positive and negative
entiation in shaping the assembly of bacterial interactions are namely illustrated by green and blue lines with
communities in activated sludge (Losos, 2008; arrows revealing an exchange or a competition for substrates or
nutrients. Mutualistic symbiosis: AOB provide nitrite (NO2-N) for
Philippot et al., 2010). NOB, which in turn remove NO2-N and thus relieve its inhibitory
Moreover, deterministic inter-taxa co-exclusion effects on AOB. Commensalism: biodegradation of macromole-
patterns are prevalent among taxonomically less cules (for example, protein hydrolysis by Saprospiraceae) into
related (or distanced) species. This phenomenon, small organic molecules, which are easily available to other
heterotrophic bacteria (OHB). Other cooperative interactions: (I)
together with our observation of almost no signifi- AOB and NOB provide NO2-N and NO3-N for denitrifying bacteria
cant negative correlations between OTUs with the (DNB), (II) HB release CO2 which is assimilated by autotrophic
same taxa, suggested that co-exclusion primarily AOB and NOB, and (III) AOB and NOB release soluble microbial
occurred between bacteria that were taxonomically products (SMPs), which are utilized by HB as carbon sources.
distanced. In a high-biomass, resource-limited Competition: (I) AOB and NOB compete with each other for
carbon sources and oxygen and with HB for oxygen and essential
biotechnical system, such as activated sludge, growth factors (EGFs); (II) different AOB, NOB or DNB compete
negative interactions are ubiquitous within or with each other for NH3-N, NO2-N or NO2-N/NO3-N, respectively;
between functional and detrimental bacteria (for and (III) different heterotrophic DNB or OHB compete with each
example, nitrifiers vs heterotrophs (Nitrosomonas other for carbon sources. A color version of this figure is available
on The ISME Journal online.
vs Clostridium XI; Nitrospira vs TM7; Figure 5); BFB
vs nutrient-removal organisms (Mycobacterium vs
Nitrosomonas; Caldilinea vs Azoarcus; Figure 5)
and floc-forming vs filamentous microbes), which, (Supplementary Figure S11b) is a typical instance
in general, reflect fierce competition between these of commensalism, that is, the former cross-feed on
bacteria for limited resources of essential growth amino acids from protein hydrolyzed by the latter, as
factors, dissolved oxygen, carbon source or other is often found in biodegradation (Faust and Raes, 2012).
substrates (Daims et al., 2006, Seviour and Nielsen, On the basis of these meaningful observations,
2010) (Figure 6). we predict that the deterministic co-occurrence
Strikingly, non-random inter-taxa co-occurrence observed between TM7 and Chloroflexi (Anaerolineae;
patterns between taxonomically distanced bacteria Supplementary Figure S11a) could be derived from
in activated sludge most likely suggest species a cooperative relation. These two types of bacteria
interactions, such as mutualism and commensalism. have been detected by FISH to co-occur in filament
For example, AOB Nitrosomonas co-occurs with epiphytic protein-hydrolyzing communities of five
NOB Nitrospira (Figure 5) out of a relation of full-scale WWTPs (Xia et al., 2007). The recent
mutualistic symbiosis, in which AOB provides construction and analysis of TM7 genomes indicate
nitrite for NOB, and in return, NOB removes nitrite that microaerophilic TM7 often buries its coccus
to prevent its inhibition on AOB. The co-occurrence cells deeply in flocs and primarily ferments glucose
between the commensal bacteria Flavobacteria and and other sugars in bioreactors (Albertsen et al.,
the protein-hydrolyzing bacteria Saprospiraceae 2013). On the basis of this knowledge, it is

The ISME Journal


speculated that TM7 may colonize to filamentous
Chloroflexi in activated sludge bioflocs because of
the benefits of minimized oxygen exposure and the
easier adsorption of organic molecules (usually in
the form of colloids and small particles) caught by
the filaments from the bulk wastewater. In return,
fermentative TM7 may provide substrates to its
filamentous host. Overall, our observation that
co-occurring bacteria tend to be taxonomically less
related essentially resembles the phylogenetic
overdispersion of co-occurring species of plants or
animals observed in many studies (Losos, 2008;
Bennett et al., 2013), revealing that negative
interactions (such as competition) have an
important impact on the community assembly of
large varieties of biological communities from
microorganisms (for example, bacteria) to macro-
scopic plants and animals.
Finally, the dominance of speciesspecies correla-
tions over those between environment and species,
as well as the lack of strong correlations between
environmental variables and many persistent OTUs,
may relate to the fact that the activated sludge is
kept operated/cultured in the artificially controlled
(thus relatively stable) Shatin WWTP, where cli-
mates show no significant seasonal variations, thus,
indicating that the variations in bacterial abundance
were primarily driven more by biological interac-
tions than by temporal changes in the physico-
chemical and operational parameters. It is also
possible that unmeasured influential variables
could exist and contribute to instances of bacterial
occurrence or changes in abundance. Among all 15
measured variables, SRT and inorganic nitrogen (for
example, NH3-N and NO3-N) in the AT best explain
partial phylogenetic and quantitative variances and
indirectly affect bacterial assembly. Generally, SRT
selects microbial populations based on their growth
rates, and thus can strongly select against slowly
growing nitrifying bacteria (especially NOB), parti-
cularly in the case of low temperatures in winter
when growth rate is lower. The positive correlations
between (I) SRT, mixed liquor suspended solids,
NO3-N, Z(NH3-N) and (II) bacterial a-diversity
indicate that appropriately extending SRT or main-
taining sufficient biomass is beneficial for improv-
ing bacterial biodiversity and ammonium removal in
activated sludge. This result, in turn, helps us to
explain or to predict how the system performance
(for example, NH3-N removal) can respond to
changes in operational conditions, considering a
close link between microbial diversity and process
robustness. From an engineers perspective, realiz-
ing that (I) maintaining rationally assembled micro-
bial community structure (in terms of both diversity
and abundance) is critical to sustaining long-term
satisfactory and steady performance and that (II) the
community structure is highly dependent on the
biological speciesspecies interactions, which can
be manipulated indirectly via the control of certain
key operational parameters (for example, SRT and
Bacterial assembly and temporal dynamics in AS
F Ju and T Zhang
693
organic loadings) and physicochemical conditions
(for example, inorganic nitrogen concentrations),
can change ways of thinking when we operate
WWTPs. For example, in the case of an incomplete
nitrification event, priority could be given to think-
ing of operational or chemical measures to inhibit
the growth of potential competing bacteria (for
example, BFB) to promote the representativeness
and effectiveness of functional nitrifying bacteria.
Overall, the fulfillment of these innovative attempts
in microbial manipulation toward better process
performance should be established by acquiring
more fundamental knowledge regarding the com-
plex interactions among microbial communities.
More studies or scientific attempts toward this
inspiring goal are warranted.
Conflict of Interest
The authors declare no conflict of interest.
Acknowledgements
We would like to thank GRF of Hong Kong for financial
support (7195/06E, 7197/08E, 7202/09E, 7198/10E, 7201/
11E and 7190/12E) on this long-term research project. FJ
would like to thank the University of Hong Kong for the
postgraduate scholarship. We would like to thank Dr Feng
GUO for his help on DNA extraction and PCR, plus Dr Lin
YE and Miss Qingmei YAN for help on sludge sampling.
We also wish to thank DSD (Drainage Services Depart-
ment) for sharing of operation data.
Author contributions
TZ designed the experiments, supervised the data
analysis and revised the final manuscript. FJ
performed the sampling, bioinformatics, statistics
and network analysis of the raw sequencing data,
wrote python scripts for data analysis and wrote the
manuscript draft.
References
Albertsen M, Hugenholtz P, Skarshewski A, Nielsen KL,
Tyson GW, Nielsen PH. (2013). Genome sequences of
rare, uncultured bacteria obtained by differential
coverage binning of multiple metagenomes. Nat
Biotechnol 31: 533538.
Barberan A, Bates ST, Casamayor EO, Fierer N. (2011).
Using network analysis to explore co-occurrence
patterns in soil microbial communities. ISME J 6:
343351.
Bennett JA, Lamb EG, Hall JC, Cardinal-McTeague WM,
Cahill JF. (2013). Increased competition does not lead
to increased phylogenetic overdispersion in a native
grassland. Ecol Lett 16: 11681176.
Bond PL, Hugenholtz P, Keller J, Blackall LL. (1995).
Bacterial community structures of phosphate-removing
and non-phosphate-removing activated sludges from
The ISME Journal
 Bacterial assembly and temporal dynamics in AS
F Ju and T Zhang
694
sequencing batch reactors. Appl Environ Microbiol 61:
19101916.
Caporaso JG, Kuczynski J, Stombaugh J, Bittinger K,
Bushman FD, Costello EK et al. (2010). QIIME allows
analysis of high-throughput community sequencing
data. Nat Methods 7: 335336.
Daims H, Taylor MW, Wagner M. (2006). Wastewater
treatment: a model system for microbial ecology.
Trends Biotechnol 24: 483489.
Edgar RC. (2010). Search and clustering orders of magnitude
faster than BLAST. Bioinformatics 26: 24602461.
Eikelboom DH. (2000). Process Control of Activated
Sludge Plants by Microscopic Investigation. IWA
Publishing: London, UK.
Eiler A, Heinrich F, Bertilsson S. (2011). Coherent
dynamics and association networks among lake
bacterioplankton taxa. ISME J 6: 330342.
Faust K, Raes J. (2012). Microbial interactions: from
networks to models. Nat Rev Microbiol 10: 538550.
Gilbert JA, Steele JA, Caporaso JG, Steinbruck L, Reeder J,
Temperton B et al. (2011). Defining seasonal marine
microbial community dynamics. ISME J 6: 298308.
Grady Jr CL, Daigger GT, Love NG, Filipe CD,
Leslie Grady C. (2011). Biological Wastewater
Treatment. IWA Publishing: London, UK.
Guo F, Zhang T. (2012). Profiling bulking and foaming
bacteria in activated sludge by high throughput
sequencing. Water Res 46: 27722782.
Guo F, Ju F, Cai L, Zhang T. (2013). Taxonomic precision of
different hypervariable regions of 16S rRNA gene
and annotation methods for functional bacterial
groups in biological wastewater treatment. PLoS One
8: e76185.
Ibarbalz FM, Figuerola EL, Erijman L. (2013). Industrial
activated sludge exhibit unique bacterial community
composition at high taxonomic ranks. Water Res 47:
38543864.
Ihaka R, Gentleman R. (1996). R: a language for data
analysis and graphics. J Comput Graph Stat 5:
299314.
Jenkins D, Richard MG, Daigger GT. (2004). Manual on the
Causes and Control of Activated Sludge Bulking,
Foaming, and Other Solids Separation Problems.
IWA Publishing: London, UK.
Ju F, Guo F, Ye L, Xia Y, Zhang T. (2013a). Metagenomic
analysis on seasonal microbial variations of activated
sludge from a full-scale wastewater treatment plant
over 4 years. Environ Microbiol Rep 6: 8089.
Ju F, Xia Y, Guo F, Wang Z, Zhang T. (2013b). Taxonomic
relatedness shapes bacterial assembly in activated
sludge of globally distributed wastewater treatment
plants. Environ Microbiol 16: 24212432.
Juretschko S, Timmermann G, Schmid M, Schleifer K-H,
Pommerening-Roser A, Koops H-P et al. (1998).
Combined molecular and conventional analyses of
nitrifying bacterium diversity in activated sludge:
Nitrosococcus mobilis and Nitrospira-like bacteria as
dominant populations. Appl Environ Microbiol 64:
30423051.
Kragelund C, Remesova Z, Nielsen JL, Thomsen TR,
Eales K, Seviour R et al. (2007). Ecophysiology of
mycolic acid-containing Actinobacteria (Mycolata) in
activated sludge foams. FEMS Microbiol Ecol 61:
174184.
Kwon S, Kim T-S, Yu GH, Jung J-H, Park H-D. (2010).
Bacterial community composition and diversity of
a full-scale integrated fixed-film activated sludge
The ISME Journal
system as investigated by pyrosequencing. J Microbiol
Biotechnol 20: 17171723.
Lipson DA, Schmidt SK. (2004). Seasonal changes in
an alpine soil bacterial community in the Colorado
Rocky Mountains. Appl Environ Microbiol 70:
28672879.
Losos JB. (2008). Phylogenetic niche conservatism,
phylogenetic signal and the relationship between
phylogenetic relatedness and ecological similarity
among species. Ecol Lett 11: 9951003.
McDonald D, Price MN, Goodrich J, Nawrocki EP,
DeSantis TZ, Probst A et al. (2012). An improved
Greengenes taxonomy with explicit ranks for ecologi-
cal and evolutionary analyses of bacteria and archaea.
ISME J 6: 610618.
McLellan S, Huse S, Mueller-Spitz S, Andreishcheva E,
Sogin M. (2010). Diversity and population structure
of sewage-derived microorganisms in wastewater
treatment plant influent. Environ Microbiol 12:
378392.
Newman MEJ. (2006). Modularity and community struc-
ture in networks. Proc Natl Acad Sci 103: 85778582.
Nielsen PH, Mielczarek AT, Kragelund C, Nielsen JL,
Saunders AM, Kong Y et al. (2010). A conceptual
ecosystem model of microbial communities in
enhanced biological phosphorus removal plants.
Water Res 44: 50705088.
Nielsen PH, Saunders AM, Hansen AA, Larsen P,
Nielsen JL. (2012). Microbial communities involved
in enhanced biological phosphorus removal from
wastewatera model system in environmental
biotechnology. Curr Opin Biotechnol 23: 452459.
Paver SF, Hayek KR, Gano KA, Fagen JR, Brown CT,
Davis-Richardson AG et al. (2013). Interactions
between specific phytoplankton and bacteria
affect lake bacterial community succession. Environ
Microbiol 15: 24892504.
Philippot L, Andersson SG, Battin TJ, Prosser JI,
Schimel JP, Whitman WB et al. (2010). The ecological
coherence of high bacterial taxonomic ranks. Nat Rev
Microbiol 8: 523529.
Quince C, Lanzen A, Davenport RJ, Turnbaugh PJ. (2011).
Removing noise from pyrosequenced amplicons. BMC
Bioinformatics 12: 38.
Reeder J, Knight R. (2010). Rapid denoising of
pyrosequencing amplicon data: exploiting the rank-
abundance distribution. Nat Methods 7: 668689.
Ruan Q, Dutta D, Schwalbach MS, Steele JA, Fuhrman JA,
Sun F. (2006). Local similarity analysis reveals
unique associations among marine bacterioplankton
species and environmental factors. Bioinformatics 22:
25322538.
Sanapareddy N, Hamp TJ, Gonzalez LC, Hilger HA,
Fodor AA, Clinton SM. (2009). Molecular diversity
of a North Carolina wastewater treatment plant as
revealed by pyrosequencing. Appl Environ Microbiol
75: 16881696.
Schmid M, Thill A, Purkhold U, Walcher M, Bottero JY,
Ginestet P et al. (2003). Characterization of activated
sludge flocs by confocal laser scanning microscopy
and image analysis. Water Res 37: 20432052.
Seviour R, Nielsen PH. (2010). Microbial Ecology of
Activated Sludge. IWA Publishing: London, UK.
Shannon P, Markiel A, Ozier O, Baliga NS, Wang JT,
Ramage D et al. (2003). Cytoscape: a software
environment for integrated models of biomolecular
interaction networks. Genome Res 13: 24982504.

Shin YK, Hiraishi A, Sugiyama J. (1993). Molecular
systematics of the genus Zoogloea and emendation of
the genus. Int J Syst Bacteriol 43: 826831.
van der Gast CJ, Walker AW, Stressmann FA, Rogers GB,
Scott P, Daniels TW et al. (2010). Partitioning core and
satellite taxa from within cystic fibrosis lung bacterial
communities. ISME J 5: 780791.
Wagner M, Loy A. (2002). Bacterial community composi-
tion and function in sewage treatment systems. Curr
Opin Biotechnol 13: 218227.
Wanner J. (1994). Activated Sludge Bulking and Foaming
Control. CRC Press: Florida, USA.
Xia L, Steele J, Cram J, Cardon Z, Simmons S, Vallino J et al.
(2011). Extended local similarity analysis (eLSA) of
microbial community and other time series data with
replicates. BMC Syst Biol 5: S15.
Xia S, Duan L, Song Y, Li J, Piceno YM, Andersen GL et al.
(2010). Bacterial community structure in geographically
Supplementary Information accompanies this paper on The ISME Journal website (http://www.nature.com/ismej)
Bacterial assembly and temporal dynamics in AS
F Ju and T Zhang
695
distributed biological wastewater treatment reactors.
Environ Sci Technol 44: 73917396.
Xia Y, Kong Y, Nielsen PH. (2007). In situ detection of
protein-hydrolysing microorganisms in activated
sludge. FEMS Microbiol Ecol 60: 156165.
Xia Y, Kong Y, Thomsen TR, Nielsen PH. (2008).
Identification and ecophysiological characterization
of epiphytic protein-hydrolyzing Saprospiraceae
(Candidatus Epiflobacter spp.) in activated sludge.
Appl Environ Microbiol 74: 22292238.
Yu K, Zhang T. (2012). Metagenomic and metatranscrip-
tomic analysis of microbial community structure
and gene expression of activated sludge. PLoS One
7: e38183.
Zhang T, Shao M-F, Ye L. (2012). 454 Pyrosequencing
reveals bacterial diversity of activated sludge
from 14 sewage treatment plants. ISME J 6:
11371147.
The ISME Journal